Research On Fault Diagnosis Technology Of Gears Based On KFCM And Information Fusion

Gear is one of the most widely used equipment to provide power. It usuallyworks in the gear box and soaks in the lubricating oil, which is affected by thecomplex working environment around it. There are usually strong nonlinear relationsbetween the fault signatures. The fault diagnosis process is influenced by manyuncertainty factors. There are some inherent uncertainties between them. Singlesignal characteristics usually can not comprehensive characterize the equipmentrunning status. So it can not ensure the accuracy of fault diagnosis. Kernel methodand multi-information fusion technology can provide a new approach to solve theseproblems such as gear fault diagnosis. This paper is begin with the review andconclusion of the research status for the diagnosis technology of the faults of gearsystem, the gear fault diagnosis experiment scheme is designed on the basic ofanalyzing the characteristic of the gear fault signal and the common fault mechanism.The gear fault diagnosis method research is carried out from two aspects:optimization improvement method of and multi-information fusion in the kernelmethod. The main work contents are as follows:1. Carried out based on the parameter optimization kernel clustering algorithmof gear fault diagnosis method research:(1) For the nonlinear characteristic of fault gear, nuclear parameters affect thesize of the problem and the result of the KFCM traditional KFCM nuclear parameteroptimization, we should study algorithms of KFCM kernel parameter optimization,the maximum and minimum distance method are applied to the KFCM kernelparameters, using the distance between the feature space of the within class distanceand the inter class,when making KFCM analysis, let the minimum distance classesfeature space, the maximum distance between the classes from the kernel parametersfrom selected to meet the requirements of the kernel function, parameteroptimization algorithm to build a nuclear model to analyze the performance of theimproved algorithm.KFCM kernel parameter optimization algorithm categoryfeature information is stronger than traditional classification features KFCM, is aneffective method of pattern recognition.(2) For non-linear features and combinations of features gear failures ofhigh-dimensional,we have established of a nuclear-based parameter optimizationKFCM gear fault diagnosis model, we can use the time domain feature extraction and feature dimensionality of the data samples, input it to the KFCM classifier,sowe can identified the gear fault classification.2. Research on fault diagnosis method based on correlation function weightedKPCA and KFCM information fusion:(1) For the problem that the result of KPCA in noise points exist principalcomponent extraction is bad, and puts forward a weighted KPCA method based oncorrelation function. In order to reduce the low credibility of data sample and theinfluence of the result of the fusion, get the result of realization of data samples ofthe weighted distribution based on means of correlation function, KPCA serialfusion, to improve the reliability of the fusion feature..(2) For the problem that a lot of uncertainty information for fault diagnosis ingear，and with a strong non-linear relationship between the characteristics of gearfault, so we use the information fusion technology can effectively deal withuncertainty information, and the KPCA and KFCM have strong nonlinear processingcapability,it established information fusion weighted KPCA and KFCM faultdiagnosis model, experimental results show that the model, while ensuring theclassification accuracy, but also has better fault tolerance.